I feel that the overly verbose AI generated marketing text on the page doesn’t do this product justice, given that the product’s claim is that it can reduce the verbosity of AI generated output.

Not sure about this. Codex already uses structured diffs, which are much simpler conceptually and shouldn’t require full file rewrites. The GPT model is probably tuned for its apply_patch tool, too. I’d like to see some real world benchmarks/comparisons.

I’m confused by the premise of this. When my AI wants to change a few lines, it changes those few lines. It doesn’t rewrite the file. This smells like snake oil.

spotify put one out https://news.ycombinator.com/item?id=49571465

snake oil